linux-kernel - x86: XSAVE flags mystery

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <Y8BDveL74VlBFZpw@thinkstation.cmpxchg8b.net>
Date:   Thu, 12 Jan 2023 09:30:37 -0800
From:   Tavis Ormandy <taviso@...il.com>
To:     linux-kernel@...r.kernel.org
Subject: x86: XSAVE flags mystery

Hello, I have a mystery and wonder if it might be a kernel bug.

I've been trying to track down why the tests in a checkpointing library
were flakey. It turns out the xstate_bv flag appears to change
non-deterministically when using AVX instructions... but why is that?

Is this a bug? Is there some state being leaked across a context switch?

Here is a reproducer:

https://godbolt.org/z/h8nY4d6c4

I would expect this to infinite loop, but if you click recompile (the
circle arrow icon at the bottom), notice the number of tests change and
it does exit the loop.

I've also attached it for reference, please compile it with gcc -mavx.

I don't know, any ideas?

Tavis.

-- 
 _o)            $ lynx lock.cmpxchg8b.com
 /\\  _o)  _o)  $ finger taviso@....org
_\_V _( ) _( )  @taviso

View attachment "xsave.c" of type "text/plain" (3038 bytes)